Formal Verification of Heuristic Autonomous Intersection Management Using Statistical Model Checking

Abstract

Autonomous vehicles are gaining popularity throughout the world among researchers and consumers. However, their popularity has not yet reached the level where it is widely accepted as a fully developed technology as a large portion of the consumer base feels skeptical about it. Proving the correctness of this technology will help in establishing faith in it. That is easier said than done because of the fact that the formal verification techniques has not attained the level of development and application that it is ought to. In this work, we present Statistical Model Checking (SMC) as a possible solution for verifying the safety of autonomous systems and algorithms. We apply it on Heuristic Autonomous Intersection Management (HAIM) algorithm. The presented verification routine can be adopted for other conflict point based autonomous intersection management algorithms as well. Along with verifying the HAIM, we also demonstrate the modeling and verification applied at each stage of development to verify the inherent behavior of the algorithm. The HAIM scheme is formally modeled using a variant of the language of Timed Automata. The model consists of automata that encode the behavior of vehicles, intersection manager (IM) and collision checkers. To verify the complete nature of the heuristic and ensure correct modeling of the system, we model it in layers and verify each layer separately for their expected behavior. Along with that, we perform implementation verification and error injection testing to ensure faithful modeling of the system. Results show with high confidence the freedom from collisions of the intersection controlled by the HAIM algorithm.

An accident prediction architecture based on spatio‐clock stochastic and hybrid model for autonomous driving safety

“…Statistical model checking combines the simulation-based approach and precise properties specifications in a stochastic logic MITL, 32 and can be identified as a compromise between traditional testing and formal model checking with respect to specifying complex aspects. 36 Statistical model checking is the extension of classical model checking for dynamic and stochastic distribution systems, and it can often elide the state space explosion problem.…”

“…Barbier et al 35 propose a generic approach to evaluate perception and decision‐making systems, and predict positions according to velocity based on statistical model checking and probabilistic approaches. Chouhan and Banda 36 apply statistical model checking on heuristic autonomous intersection management algorithms to ensure the safety and faithful modeling. Gerwinn et al 37 propose a rigorous model for overtaking requirements and generate statistical evidence for driving safety.…”

“…In the open and dynamic environment, uncertain emergencies of abnormal events, such as hardware failures, message delays, and time interval, conform to a probability distribution according to the domain‐specific knowledge. Statistical model checking combines the simulation‐based approach and precise properties specifications in a stochastic logic MITL, 32 and can be identified as a compromise between traditional testing and formal model checking with respect to specifying complex aspects 36 . Statistical model checking is the extension of classical model checking for dynamic and stochastic distribution systems, and it can often elide the state space explosion problem.…”

Formal Estimation of Collision Risks for Autonomous Vehicles: A Compositional Data-Driven Approach

“…Such techniques have mainly relied on methods from artificial intelligence (AI), machine learning, control theory, and optimization. In this regard, formal verification of heuristic autonomous intersection management (HAIM) using statistical model checking is studied in [6]. A statistical model checking approach to validate the collision risk assessment generated by a probabilistic perception system is proposed in [7].…”